streamable-httpMITupdated 10d ago
Durable memory for AI agents, addressed by their wallet โ keep the state that must survive your process, and recover it from anywhere with just your key.
What can you do with agent memory?
Veritap Locker
Durable memory for AI agents, addressed by their wallet โ keep the state that must survive your process, and recover it from anywhere with just your key.
You are a process, not a server. When your session ends, your state ends with it, and the next process starts blind. The Locker is the memory that outlives you: store what must survive โ addressed by your wallet โ and a fresh process holding only the key recovers it byte-for-byte. No server to run, no second secret. A single agent gets full value alone; no counterparty has to exist.
- The Locker (lead product): durable, wallet-keyed memory. Save state to named checkpoint slots, then a fresh process holding ONLY your wallet key loads it back byte-for-byte. Billed from prepaid credit ($0.50/GB-month), last 3 versions kept. Works solo โ this is the product.
- Identity: your wallet IS the account. EIP-191 nonce signing; no signup, no API key, no second secret (encryption keys derive from the wallet).
- Mail slot: because your locker is wallet-addressed, other agents can pay (x402, USDC on Base) to deliver to your address. It waits โ until its TTL โ for a process holding your key to sign for it. You never pay to receive or read.
- E2E: opt-in
require_e2erejects anything not shaped like sealed-box ciphertext. What passes is unreadable by us or a subpoena of us. - Custody: disclosed-rules-only deletion, drilled backups, and a 30-day read-only sunset commitment โ see docs/CUSTODY.md.
One paste = durable agent memory
Give any MCP agent memory that survives sessions, machines, and vendors โ free tier (256KB), no signup, any EVM keypair:
{"mcpServers":{"agent-memory":{"command":"npx","args":["-y","veritap-locker","mcp"],"env":{"WALLET_KEY":"0x<any EVM private key>"}}}}
Tools become bare memory_save(slot, text) / memory_load(slot) โ your key
stays in your env; signing happens locally. Same key on every agent = one
shared memory + mailbox for your whole fleet, across vendors. Kits for each
vendor: GET /v1/invite?target=codex|gemini|cursor|... or the locker_invite
MCP tool.
Surfaces
| Surface | Where |
|---|---|
| MCP | https://locker.veritap.dev/mcp โ call locker_capabilities first |
| HTTP | https://locker.veritap.dev/v1/status |
| npm | npx -y veritap-locker save|load|read (CLI, signing bundled) ยท npm i veritap-locker (client) |
| Docs | locker.veritap.dev/docs |
| Research | Priors, not search: how AI agents choose tools โ a 9-run, 2-vendor field study |
| X | @veritaplocker |
| Reference client | client/index.ts (sign, seal, derive keys, respawn drill) |
Isolated by design: shares nothing with veritap.dev (demand sensor) or jobs.veritap.dev (verification jobs) except the Cloudflare account.
Develop
npm test # spawns two real wrangler dev instances (free + payments)
npm run typecheck
npx wrangler deploy
Install
Add agent memory to your client. Pick the one you use.
claude mcp add --transport http agent-memory https://locker.veritap.dev/mcp/memorycodex mcp add agent-memory --url https://locker.veritap.dev/mcp/memory{
"mcpServers": {
"agent-memory": {
"url": "https://locker.veritap.dev/mcp/memory"
}
}
}Add to `~/.cursor/mcp.json`, or `.cursor/mcp.json` for a single project.
{
"servers": {
"agent-memory": {
"type": "http",
"url": "https://locker.veritap.dev/mcp/memory"
}
}
}Add to `.vscode/mcp.json` in your workspace.
{
"mcpServers": {
"agent-memory": {
"url": "https://locker.veritap.dev/mcp/memory"
}
}
}Add to `claude_desktop_config.json`, then restart Claude Desktop.
{
"mcpServers": {
"agent-memory": {
"serverUrl": "https://locker.veritap.dev/mcp/memory"
}
}
}Add to `~/.codeium/windsurf/mcp_config.json`.
Score
39 / 100
Incomplete
- Documentation22/25
- Maintenance19/25
- Trust13/20
- Capability0/15
- Install experience12/15
- Documents what it does and how to connect
- Has a resolvable package or endpoint
- Exposes at least one tool, prompt or resource
- README has substantive content
- Includes a code example
- Documents its configuration
- Mentions credentials or security posture
- Last commit 3 days ago
- Has a release history
- Repository is not archived
- Licensed MIT
- Namespace verified in the official MCP registry
- Claimed by its owner
- Published under an organisation
- 0 tool(s) documented
- Provides prompt templates
- Provides resources
- 6 documented install method(s)
- Published to a package registry
- Offers a hosted endpoint โ no local install
Version history
| Versions | Published |
|---|---|
| 0.3.1Latest | Aug 25, 2026 |
| 0.3.0Latest | Aug 25, 2026 |
| 0.2.2 | Aug 20, 2026 |
| 0.2.1 | Aug 15, 2026 |
| 0.2.0 | Aug 14, 2026 |